configuring-sixth-senselisted

# Setup Sixth Sense Run this skill once per project to configure how Sixth Sense dispatches subagents. ## Interview Ask the user the following questions, one at a time. For each question, provide a recommended answer: Note: Opus can run two distinct ways, and they don't share a toggle. The narrow second-opinion pass (Q4 below) is read-only, always separate from dispatch, on by default per benchmark evidence. `opus-high` (Q5 below) is a real Effort Routing tier for genuinely architectural/planning work and the fix loop's last-resort escalation — it participates in normal dispatch the way `sonnet-high` does, just reserved for when that isn't enough. ### 1. Authorized models > Which models can subagents use? > - Sonnet only (no Haiku batching — everything, including mechanical tasks, runs on Sonnet) > - Sonnet + Haiku (recommended — trivial, well-specified tasks batch onto Haiku for a fraction of the cost; see Q3) ### 2. Default effort for Sonnet > What effort level should Sonnet use for standard work? > - **low** — well-defined tasks, low ambiguity risk > - **medium** (recommended) — good tradeoff for everyday work > - **high** — complex tasks, architectural decisions ### 3. Haiku batching > Trivial tasks (single-file change, <20 lines, no architectural decision) are batched onto Haiku **by default** — the eligibility bar is already narrow enough to act as a guardrail. Turn this off only if you want every task on Sonnet regardless of cost. > - **Keep it on** (default, re